Color Image Hashing Algorithms Based On Quaternion SVD And Sparse Model

Image hashing is a useful technique of digital media research.It is a cross-research topic in the field of image processing and information security,which has been successfully used in image retrieval,image forensics,image authentication,image copy detection,image watermarking and image quality assessment.Image hashing algorithm can calculate the compressed representation of the input image and replace it with the image itself,thus effectively reducing the image storage cost.It maps visually identical images to the same or similar hashes based on visual content of images.The first property of image hashing algorithm is called robustness.The demand of this property is that image hashing algorithm should be robust to normal digital operations,such as JPEG compression and image enhancement.This is because after these digital operations,the processed images are still similar with their original versions,but their digital representations are quite different.The second property of image hashing algorithm is called discrimination.It requires that image hashing algorithm should produce different hashes for different images.This property should effectively distinguish images with different contents since there are many different images in practical applications.In fact,robustness and discrimination are two important properties of image hashing algorithm.Improving the classification performance between robustness and discrimination is an important task in current image hashing research.This paper exploits the theory and technology of quaternion singular value decomposition(QSVD),visual attention model,color vector angle(CVA)and sparse model to investigate image hashing algorithm,and designs two efficient and novel image hashing algorithms,which improves classification performance of hashing algorithm in robustness and discrimination.The first image hashing algorithm is based on quaternion singular value decomposition.The second image hashing algorithm is based on sparse model.The primary research results are summarized as follows.1.Image hashing algorithm based on quaternion singular value decomposition is proposedQuaternion is a useful mathematical theory.When the quaternion is used to represent color image,it can process color image in a holistic manner without discarding color information.Currently,quaternion has been widely used in handling color images,such as color image de-noising,color image registration and color image watermarking.In this paper,we exploit QSVD to study color image hashing algorithm and propose construct the image hash with quaternion singular values.Specifically,the proposed hashing algorithm first uses QSVD to extract the quaternion singular values of image block as features from the CIE L~*a~*b~*color space and views each image features of a block as a point in the Cartesian coordinates,and compresses the features by calculating the Euclidean distance between its point and a reference point.Finally,image hash is formed by concatenating all Euclidean distances.Experiments with three open image databases are conducted to validate efficiency of this image hashing algorithm.The results demonstrate that this image hashing algorithm can resist many digital operations and reaches a good discrimination.2.Image hashing algorithm based on sparse model is proposedSparse model is a useful method for high-dimensional data processing and has been widely used in fields such as image processing and pattern recognition.In this paper,we design a new image hashing algorithm based on Itti visual attention model and sparse model.This algorithm uses CVA and Itti visual attention model to construct image color-weighted representation,from which feature extraction can effectively improve robustness.Next,a sparse model called robust principal component analysis is used to decompose the color-weighted representation into a low rank component and a sparse component.Because low rank component can reflect internal structure of the original image,using low rank component to construct the image hash ensures that the algorithm has good discrimination.The results demonstrate that this hashing algorithm reaches good robustness and discrimination for many digital operations and can be applied to image copy detection.Receiver operating characteristic(ROC)curve is used to analyze classification performance of the two hashing algorithms proposed in this paper in robustness and discrimination,and the performance is compared with several existing hashing algorithms.Experimental results show that the proposed image hashing algorithms outperform the compared algorithms in classification performance.